Engineering for decisions that cannot fail.
Cognitive — Deep Tech
We design, implement and scale.
Artificial intelligence systems for governments and large organizations across Latin America and Europe.
03 — Track record
National government · Argentina
City government · Argentina
Municipal government · Tres de Febrero, Buenos Aires
Commercial aviation
04
Countries with commercial presence: Argentina, Spain, Mexico and Paraguay.
Applied research agreement with Universidad Tecnológica Nacional.
04 — Capabilities · 1/2
Cognitive Units
Six lines of work, from strategy to applied research.
- 01
Strategic AI
Custom AI aligned with institutional goals: from idea to production.
Design and implementation of artificial intelligence systems aligned with strategic, institutional and business goals. AI built for leaders, not only for systems.
Applications
- Assessment and strategy
- Custom models
- Applied generative AI
- 02
Computer Vision & Analytics
Images and video turned into real-time decisions.
Computer vision for security, retail, industry, agriculture, government and critical infrastructure, integrated with the cameras and systems the client already runs.
Applications
- Security
- Retail
- Industry
- Agriculture
- Government
- Critical infrastructure
- 03
Automation & Agents
AI agents that run complex processes and decide in real time.
Intelligent automation built on agents that execute processes, coordinate workflows and make real-time decisions, combining agents, conversational AI and automation.
Applications
- Operations
- Customer service
- Sales
- Back office
04 — Capabilities · 2/2
Cognitive Units
- 04
Conversational AI & Avatars
Conversational systems and human avatars, aligned with brand identity.
Advanced conversational assistants and human avatars for marketing, sales, customer service and institutional communication, available around the clock.
Applications
- Marketing
- Sales
- 24/7 support
- Institutional communication
- 05
Blockchain & AI
Traceability, security and governance for critical, regulated environments.
Architectures that combine artificial intelligence and blockchain to guarantee traceability, security and governance in critical and regulated environments.
Applications
- Traceability
- Security
- Data governance
- 06
CGN Labs
Applied research with Universidad Tecnológica Nacional.
R&D in quantum AI, new computational models, self-sovereign identity (SSI) and emerging technologies, under an agreement with UTN.
Applications
- Quantum AI
- Self-sovereign identity (SSI)
- New computational models
05 — Flagship cases
AI applied to real-world challenges.
We design solutions that connect data, models and operations: from computer vision, geospatial analysis and drones to automation and generative AI.
05 — Flagship cases · 01 / 06
The city, seen in motion
Computer vision and geolocation that turn urban routes into actionable information.
We process images captured from vehicles to identify public road conditions and place them on the map. The result combines visual evidence with georeferenced data to help prioritize surveys, maintenance and operational response.
- Technology
- Computer vision · Geolocation · Georeferenced data
- Applications
- Surveys · Maintenance · Operational response
05 — Flagship cases · 02 / 06
Visual intelligence for Retail Shelf
From a walk down the aisle to an intelligent reading of the retail shelf.
We combine image capture, visual recognition and text extraction to analyze products, facings and labels on the shelf. We turn point-of-sale images into structured information for auditing, retail execution and assortment analysis.
- Technology
- Image capture · Visual recognition · Text extraction
- Applications
- Auditing · Retail execution · Assortment analysis
05 — Flagship cases · 03 / 06
GeoAI for land, environment and agriculture
Satellites, GIS and artificial intelligence to detect patterns at a territorial scale.
We integrate satellite imagery, geographic data and vision models to identify areas of interest across large expanses. From detecting open-air dumps to analyzing land cover and productive zones, we turn geospatial information into maps that help guide research and decision-making.
- Technology
- Satellite imagery · GIS · Vision models
- Applications
- Open-air dumps · Land-cover analysis · Productive zones
05 — Flagship cases · 04 / 06
Aerial intelligence with drones
Aerial capture and computer vision to observe, map and analyze from new perspectives.
We combine aerial imagery with computer vision tools and geospatial analysis to build a detailed view of the territory. Drone workflows can be adapted to surveys, inspections and monitoring of hard-to-reach areas, integrating the captured data with maps and analytical models.
- Technology
- Aerial imagery · Computer vision · Geospatial analysis
- Applications
- Surveys · Inspections · Hard-to-reach areas
05 — Flagship cases · 05 / 06
Smart urban waste management
Computer vision to support the collection and municipal management of urban solid waste.
We analyze images from collection routes to detect scattered waste and other relevant conditions in public space. Visual information can help municipalities and operators monitor services, identify areas that need attention and improve the traceability of the operation.
- Technology
- Computer vision · Image analysis
- Applications
- Service monitoring · Areas that need attention · Traceability
05 — Flagship cases · 06 / 06
Smart inspection: quality and anomalies
Computer vision to verify products, detect deviations and support quality control.
We apply visual analysis to industrial checks such as counting, fill level, labeling and anomaly detection. The models help identify variations and raise review signals, integrating automated inspection with the quality criteria of each process.
- Technology
- Computer vision · Visual analysis
- Applications
- Counting · Fill level · Labeling · Anomaly detection
06 — Engineering
How a Cognitive system fits in.
We build on the infrastructure the client already has. The path from camera to management system is the same in security, logistics, retail and agriculture.
- 01
Capture
Existing CCTV, drones, phones or robots.
- 02
Ingestion
Stream capture and image preprocessing.
- 03
Model
Neural networks custom-trained on the client’s labeled data.
- 04
Inference
In the cloud or on the client’s premises, on Docker and Kubernetes.
- 05
Event
Real-time alerts and notifications.
- 06
Integration
Data flowing into the client’s WMS, management system and dashboards.
Technology stack · 1/3
Foundation Models, Agentic Architectures and RAG
- Orchestration and agents
- LangGraphStateful Agent Workflows
- LlamaIndex
- Semantic Router
- Function Calling / Tool Useprotocols
- Advanced RAG systems
- Hybrid SearchDense + Sparse/BM25
- Cross-Encoder Re-ranking
- Contextual Chunking & GraphRAG
- Evaluation, observability and guardrails
- Langfuse
- Arize Phoenix
- Ragas
- NeMo Guardrails
- Llama-Guard
Computer Vision & Multimodal Perception
- Models and frameworks
- PyTorch
- Vision Transformers (ViTs)
- YOLO
- Segment AnythingSAM 2 and 3
- Multimodal and zero-shot embeddings
- CLIP / SigLIP
- DINOv2
- OWLv2Zero-shot detection
- Perception pipeline
- OpenCV
- Albumentations
- ONNX Runtime / TensorRToptimization and accelerated inference
- Document AI & OCR
- PaddleOCR
- Surya / Marker
- Donut
- LayoutLM
Technology stack · 2/3
Data Platforms, Vector Search & Geospatial
- Vector and hybrid databases
- Qdrant
- Milvus
- pgvectorPostgreSQL
- Redis Stack
- Data ingestion and processing
- Polars
- DuckDB
- Apache Arrow
- dbt
- Raydistributed computing for ML
- GeoAI and remote sensing
- PostGIS
- GDAL/OGR
- GeoPandas
- Rasterio
- Xarray
- Sentinel-2 / Copernicusmultispectral data
- GeoJSON
MLOps, Inference and High-Performance Deployment
- Accelerated inference servers
- vLLM
- Triton Inference Server
- TGIText Generation Inference
- Ollama / TensorRT-LLM
- Cloud & Enterprise AI
- AWSBedrock, SageMaker
- GCPVertex AI
- Hybrid / On-Premise GPU Clusters
- Infrastructure and containers
- Docker
- KubernetesK8s
- Helm
- TerraformIaC
- CI/CD and lifecycle
- GitHub Actions
- MLflow
- DVCData Version Control
- Prometheus / Grafana
Technology stack · 3/3
Enterprise Automation and BI
- Connected automation
- n8nSelf-hosted / Enterprise
- Power Automate
- Custom Python Microservices
- Event-driven Webhooks
- Analytics and Business Intelligence
- Power BI
- Streamlit / FastUIrapid prototyping of AI interfaces
CGN Labs
Applied research with UTN.
R&D in quantum AI, new computational models and self-sovereign identity (SSI), under an agreement with Universidad Tecnológica Nacional.
That work produced ThreatSpotterAI: real-time detection of violent situations in video.
07 — The organization
Engineering, data science and research in one team.
- 01
We start with the problem.
Before talking about models, we understand the operation, the available data and the decision that needs to be made. It is the first phase of CRISP-DM, and the one that shapes every other.
- 02
We build with the client’s team.
Models are trained alongside the people who know the ground. In rice field monitoring, side by side with the client’s agronomists.
- 03
We integrate before we replace.
Existing cameras, servers and systems are the starting point. Intelligence connects to what already works.
Commercial presence
- Argentina
- Spain
- Mexico
- Paraguay
08 — Credentials
Clients and partners
Public sector
Companies
Partners
LOGER
Vision Fractal
If the problem is complex, it’s our kind of problem.
Let’s talk.